Nous: Hermes 4 70B

Nous: Hermes 4 70B is a cutting-edge hybrid reasoning model developed by Nous Research, leveraging the powerful Meta-Llama-3.1-70B architecture. This model introduces a unique hybrid mode, allowing it to either respond directly or generate explicit <think>...</think> reasoning traces before providing an answer. Users can easily control this reasoning behavior using the `reasoning` `enabled` boolean parameter, providing flexibility for various applications. This 70B variant has been meticulously trained with an expanded post-training corpus of approximately 60 billion tokens, with a strong emphasis on verified reasoning data. This extensive training has led to significant improvements across critical domains such as mathematics, coding, STEM, logic, and structured outputs, all while maintaining its high performance as a general assistant. It supports advanced features like JSON mode, schema adherence, function calling, and tool use, making it highly versatile for complex tasks. Furthermore, it is engineered for greater steerability and exhibits reduced refusal rates, ensuring more reliable and consistent interactions. Key specifications include a generous context window of 131K tokens and a maximum output of 4K tokens. Pricing is competitive at $0.11 per 1M input tokens and $0.38 per 1M output tokens. It offers capabilities such as functions, code generation, streaming, and search, making it ideal for chat applications. Please note, it does not support image generation. Access tier: STARTER.